“my my take on the the uh uh the GPU uh depreciation debate is that it's nonsense, right? It's a debate that is being uh brought to the forefront by uh some…”
Intrator directly fades the GPU-depreciation short by naming traders shorting CRWV and asserting his A100s appreciated this year while clients sign five-year contracts. He is taking the long side of the exact debate Michael Burry built his CoreWeave short around.
“We signed a $9.7 billion contract with them late last year, but as I was explaining to you before the show, that's 5% of our capacity. So things are busy at…”
A $9.7B Microsoft contract is only 5% of IREN's capacity, signaling MSFT's relentless compute appetite
“Apple seems to be MIA. They don't seem to want to play.”
Apple is MIA on AI capex and ships too slowly to keep pace in the agentic era
“Memory is a throttle right now, right? Oh yeah, it is... the demand is through the roof. That's number one. Number two is is that there was probably an…”
Intrator says memory is the binding constraint on AI compute: demand is through the roof while fab capacity that should have been built in 2023 never arrived. Tight DRAM supply against surging demand is the pricing-power setup that flows straight to a pure-play memory maker like Micron.
“I I think you have to. I think the reality is it's going to take a decade... But now is the time to start that conversation, put in place policies, mobilize…”
Nuclear is inevitable for data center power; now is the time to mobilize capital on SMRs
“Or this new Dell that's coming out. And I I really think the idea of spending $10,000 on a powerful desktop will appeal to people if it lowers their $500 a…”
Calacanis names the new Dell-Nvidia AI workstation specifically and argues the $10k local rig pays for itself by cutting cloud-model bills while adding privacy. That is a demand call on the desktop AI category Dell is shipping, expressed through DELL stock.
“We have Kimmy, we have Nimatron, and we have uh a lot of usage of Quen, Alibaba Quen. Silently under the hood.”
Alibaba's Qwen quietly racking up heavy usage under the hood at orchestration layers
“I think it it happened with Opus 45. The Anthropic 45. That was an inflection point when models were started being amazingly good at orchestration and…”
Opus 4.5 was the inflection point for agentic orchestration; Claude Code seeded the whole sandbox paradigm
“And I said I made a bet publicly on the bank. I said, I am buying a bunch of Google because I believe even though they're behind, it's because they're too…”
Calacanis recounts a public bet that he bought a lot of Google on the view that it was behind only because it was too precious to ship, and that pressure to release more would close the gap. The catalyst he cites is Sergey pushing the team to release the good stuff.
“It's right up the end of the spectrum. It's gangbusters. We we cannot meet demand. That's why the whole industry now is around time to compute. There are no…”
Roberts stakes that AI compute demand is at the extreme end of the spectrum with zero idle GPUs, framing the $9.7B Microsoft deal as only 5% of IREN's 4.5GW capacity. The constraint he names is time-to-build, not demand, which is a bullish stake on monetizing locked-up power.
“there's multiple Silicon looking for homes. I think I think it's fair to say Nvidia has a massive head start. The ecosystem they've incubated, the standards…”
Roberts argues that amid a flood of custom AI silicon looking for sockets, Nvidia's incubated CUDA ecosystem and de facto standards make it the lowest-risk way to build at scale. That is a direct bull case for Nvidia's moat, which the equity is the cleanest way to own.